How much to expect from AI

Introduction

Last month I recounted my story of asking ChatGPT for help with a calculation: Don't expect too much from AI, after all, it's not only human... When the link to this article was included in the Excel Community update email the title had been abbreviated to: 'Don't expect much from AI' which is pretty much the opposite of what I believe. I like to think I've got a reasonable record on predictions: not only did I realise the potential for the Internet to be quite significant, but I also tipped Blondie for stardom from the first time I heard 'Denis Denis'. In fact, over 7 years ago, the Excel Community predicted the inclusion of AI in Excel: Microsoft April update introduces A1 for Excel.

Consequently, this article sets out my thoughts on what we can actually expect from IT and, at the risk of destroying any jeopardy at such an early stage, it is much.

Expect the unexpected

Perhaps the least contentious prediction about AI is that most of the current predictions will be wrong, not necessarily in the assessment of the importance of AI, but in the specific outcomes. For example, one of the fears about AI that has been raised most frequently relates to its ability to generate convincing fake content, from Spam emails to 'deepfake' videos of political leaders. The obvious consequence of this would be an increased incidence of people being duped by this faked content. However, it is also possible that the effect will be exactly the opposite. As convincing fakes become more prevalent, people might just become much more sceptical about relying on any content without seeking out corroborating evidence, making fake content less, rather than more, likely to mislead.

In terms of how much to expect from AI, many years ago I wondered whether the impact of the Internet was important enough to be considered as a stage in human evolution. In the end, I decided that it was more likely that humanity was a stage in the evolution of the Internet. At this point, it's hard not to think of Douglas Adams, Deep Thought and the earth as an experiment to discover the answer to life the universe and everything. If I was a pan-dimensional being seeking to harness the combined power of billions of brains, together with all the knowledge that has ever been recorded, AI would definitely be on my list of likely technologies. More immediately, we are already seeing AI change the way that Internet searches work, making enormous advances in medical technology and even helping plan our holidays.

Closer to home, or at least to the Excel community, not only is Microsoft already previewing its own AI-powered 'CoPilot' as part of the Microsoft 365 applications, but third parties are already providing AI-based Add-Ins for Excel. Most of these are designed to help create complex formulas from simple text instructions.

Going back to my original article, an Excel Community member, Gary Knott, pointed out that ChatGPT was not optimised for calculations and suggested the use of the Wolfram Alpha plug-in for ChatGPT. As far as I can see, this does require the use of the premium version of ChatGPT, ChatGPT Plus, but does offer some very significant additional capabilities (see the article: ChatGPT Gets Its "Wolfram Superpowers"!)

The ability to direct AI to use very specific resources to answer particular types of questions opens up some very interesting possibilities. In many organisations, the most usual way to solve Excel problems is to ask the office Excel expert or, failing that, to use an Internet search to provide possible solutions. AI could encapsulate the knowledge of all the Excel experts whose content is available to it and combine it with specific financial expertise. The result within Excel would not just be a suggestion as to how to address a particular problem, but a complete solution with all the calculations and formulas built and documented.

Conclusion

So, in short, I do expect much from AI, but my advice would still be to not expect too much at this stage. As ChatGPT makes very clear in its disclaimers, some answers will contain errors and there are restrictions on the data to which it has access. However, as AI evolves and becomes more capable of working with specialised information, it will quickly become increasingly accurate and reliable.

Finally, given the controversy arising from the headline, I thought the sensible thing to do was to seek the advice of ChatGPT itself. So, I pasted the text of the original article into the chat window and set it the following task:

Please generate 5 possible titles for this article:

ChatGPT